JPH1166317A - Device and method for detecting eye position of face image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an eye position detector and a method which use histogram and effectively detect the positions of eyes in a face image also with high reliability. SOLUTION: This device includes a window creating part 120 which generates such a window as to encircle a face image, a histogram extracting part 130 which extracts a histogram from the face image, a peak detecting part 140 which detects peaks from the histogram and compares the number of detected peaks with a prescribed number, an area calculating part 150 which calculates the area of a histogram to each face image when both numbers are coincident and a selecting part 170 which selects the largest area among each area and decides the positions of eyes by using the largest area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and a method for detecting the position of an eye in an image of a human face, and more particularly to an apparatus and a method for detecting an eye position in a face image using a histogram. Regarding the method.

[0002]

2. Description of the Related Art As is well known, facial images can be distinguished by features such as eyes, nose, mouth, and facial contour. Of these features, the eyes have the most important features in the face. However, accurately detecting the position of the eye on the video is a very difficult task.

In order to solve such a problem, an eye position detecting system using infrared rays has been proposed.

FIG. 1 is a schematic block diagram of a conventional eye position detection system 10, which is disclosed in US Pat.
No. 427, EYE POSITION DET
ECTION SYSTEM AND METHOD
It is disclosed under the name THEREFOR. The eye position detection system 10 for detecting the position of the eyes of a driver or the like of a car includes an infrared strobe 12, a TV camera 14,
An A / D converter 16, an image memory 18, an eye window detecting unit 20, an iris detecting unit 22, a driver state determining unit 24, and a timing circuit 26 are provided. In the system 10, the infrared strobe 12 emits infrared radiation to the driver's face. The TV camera 14 captures an image of a face irradiated by infrared rays. The timing circuit 26 matches the timing of the infrared light emitted from the infrared strobe 12 with the timing of the face image captured by the TV camera. The A / D converter 16 converts an analog signal (analog face signal) of the face image obtained by the TV camera 14 into a digital signal (digital face signal) of the face image. The image memory 18 receives the digital face signal from the A / D converter 16 and stores it. The eye window detecting section 20 detects an area including two eyes based on the video signal stored in the image memory 18. The iris detection unit 22 processes an image signal in the area detected by the eye window detection unit 20 to detect an iris position. The driver state determination unit 24 determines, for example, whether the driver is dozing or looking at another place based on the result detected by the iris detection unit 22.

However, in the conventional eye position detection system 10, the infrared strobe 1 that emits infrared light is used.
2 is required, the structure of the system is complicated, and furthermore, since the video data needs to be binarized in order to detect the position of the eyes, the detection process is complicated.

[0006]

SUMMARY OF THE INVENTION Accordingly, a main object of the present invention is to provide an eye position detecting apparatus and an eye position detecting method for effectively and highly reliably detecting an eye position in a face image using a histogram. To provide.

[0007]

According to the present invention, there is provided, in accordance with the present invention, an eye in a face image, which is an image of a human face, represented by a plurality of pixels each having a gray level value. An eye position detecting device for detecting a position of a face image, a window generating means for generating a window surrounding the face image, a histogram extracting means for extracting a histogram from the window face image, and detecting a peak from the histogram. A peak detection unit that compares the number of detected peaks with a predetermined number; and, when the number of peaks matches the predetermined number, a peak in a histogram for each of the window face images. Area calculation means for calculating an area that is the number of pixels in the range between the lowest pixel value or 0 pixel value point including the value,
And a maximum area selecting unit for selecting a maximum area from the calculated areas and determining an eye position using the maximum area.

[0008]

Preferred embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 2 shows an eye position detecting apparatus 10 according to the present invention.
0 is a schematic block diagram of the apparatus, which includes a filter 110, a window generator 120, a histogram extractor 130, a peak detector 140, an area calculator 150,
It comprises a switch 160 and a selector 170.

The present invention mainly relates to a face image, and the face image is captured by, for example, a TV camera (not shown). FIG. 3 shows a face image 200 as an example, which includes 520 pixels in the X-axis direction and 500 pixels in the Y-axis direction. The angle of the field of the face image 200 is adjusted so that the entire face just fits in the face image 200. Before the face image 200 is input to the filter 110 as an input face signal, the face image 200 is A / D
It is converted into a digital face video signal by a converter (not shown).

The filter 110 pre-processes the received digital face image signal and supplies it to the next processor. More specifically, the filter 110 removes noise included in the input video signal, for example, impulse noise, and generates a filtered video signal. To remove a very small amount of noise, a cross-shaped median filter may be used. Here, the filtering process is performed in pixel units,
Also, various known low-pass filters may be used for noise removal.

[0012] The filtered image from the filter 110 is supplied to the window generator 120. The window generator 120 sequentially generates a plurality of search windows in the filtered video. In the preferred embodiment of the present invention, search window 220 is rectangular as shown in FIG. The window generating unit 120 performs a raster scanning method.
Using g method), the entire filtered image is scanned, and a position vector PV representing the position of each search window is supplied to the switch 160 via the line L10.

The histogram extracting unit 130 extracts a histogram of a filtered image included in each search window generated by the window generating unit 120. The histogram of the search window is determined by plotting the frequency of the filtered image pixels within the search window as a function of the gray level value. Also, the histogram extracting unit 130
Calculates the frequency for each gray level value by counting the number of pixels in the search window when the gray level value of each pixel is 0 to 255. FIG. 5 shows an example of a histogram according to the present invention. In this histogram, the horizontal axis represents gray levels, and the vertical axis represents the frequency of pixels having each gray level in the search window.

Thereafter, the histogram extracting section 130 extracts the histogram and supplies the histogram to the peak detecting section 140. Based on the received histogram, the peak detection unit 140 detects, for example, three peaks P1, P2, and P3 in the gray level region of 0 to 255 as shown in FIG. Prior to detecting these peaks, the histogram is smoothed, i.e., the values of the histogram are averaged to remove small fluctuations on the histogram. The peak on the histogram is the peak point (eg, P
1) It may be defined as a gray level cluster surrounding 1) and is a local maximum of the histogram, the frequency corresponding to the peak of which is greater than a predetermined threshold. As shown in FIG. 5, a predetermined threshold value TH
May be determined as the average frequency of each pixel in the histogram. That is, a peak represents a cluster of gray levels that appears more frequently in the input video than other gray levels.

[0015] As shown in FIG.
A search window (eg, 220) surrounding or covering pupil 223 and white eye 221 is detected as a candidate window. This candidate window is white eyes 22
1. Different contrasts between the skin 222 and the pupil 223 can be divided into three regions. Therefore, when the number of peaks detected by the peak detecting unit 140 is 3, the peak detecting unit 140 determines the search window 220 having three peaks as a candidate window, and stores the histogram and peak information of the candidate window in the area. A switch signal, which is supplied to the calculation unit 150 and indicates that the search window is the candidate window 220, is supplied to the switch 160 via the line L20. Here, the peak information has a pixel value corresponding to the peak in the histogram. The switch 160 performs a connection between the window generation unit 120 and the selection unit 170 according to the switch signal from the peak detection unit 140 so that the selection unit 170 is supplied with the position vector corresponding to the candidate window.

Subsequently, as shown in FIG. 5, the area calculation unit 150 first starts with two peaks (for example, the first peak P ₁ having the largest pixel value) among the three peaks on the histogram for the candidate window. And the second peak P ₂ ) having the minimum pixel value. Then, the area calculation unit 150 calculates an area A ₁ having a first peak P ₁ and the area A ₃ having a third peak P _3. Area A ₁ represents the number of pixels that exist between 0 to L _1, the area A ₃ is L
Represents the number of pixels existing between _{2 and} 255. L ₁ corresponds to a pixel value having the smallest frequency between P ₁ and P _2, L ₂ corresponds to a pixel value having the smallest frequency between P ₂ and P _3. After that, the area calculation unit 150 calculates the sum area A _T of the two areas A ₁ and A ₃ and supplies the sum area A _T to the selection unit 170 via the line L50. Here, area A ₁ represents the number of pixels in pupil 223, and area A ₃ represents the number of pixels in white eye 221.

The selection unit 170 selects the maximum sum area of the sum area A _t to all candidate window obtained from the input face image, the position vector PV corresponding to the candidate window having the sum area of the maximum Is output as the eye position of the input face image.

FIGS. 6 and 7 are flow charts for explaining the eye position determining method according to the present invention.

The face image input from the TV camera is input to the eye position detecting device 100 (step S10), and the face image is filtered by the filter 10 (step S20). It is input to the generation unit 120 and scans the entire filtered image by the raster scanning method, and sequentially generates a search window and a position vector PV of the search window (step S30). In step S40, histogram extracting unit 1
30 extracts a histogram from the filtered video in each search window. Thereafter, a peak is detected on the histogram by the peak detection unit 140 (Step S50).

In step S60, peak detector 140
It is determined whether the number of peaks in the histogram is three or not. If the result of the determination is true, the corresponding search window is determined as a candidate window, and the process proceeds to step S70; Proceed to step S100.

In step S70, area calculation section 150
Detects two peak points corresponding to the maximum and minimum pixel values among the three peak points for each candidate window. In step S80, the area calculation unit 150 calculates the sum area A _T , that is, the total number of pixels in the peak including the two peak points obtained in step S70. In step S90, the sum area A _T and the position vector PV for each candidate window are stored in the selection unit 170,
The process proceeds to step S100.

In step S100, the process determines whether all search windows for all face images have been processed. If all search windows have not been processed, the process returns to step S40. If all the processes have been performed, the selection unit 170 selects the largest sum area among the sum areas A _T of the candidate windows for all face images (step S120). Subsequently, step S130
At, the selection unit 170 outputs a position vector corresponding to the candidate window having the largest sum area A _T , and the process ends.

While the preferred embodiment of the present invention has been described above, those skilled in the art will be able to make various modifications without departing from the scope of the present invention.

[0024]

Therefore, according to the present invention, there is no need to provide an infrared strobe for radiating infrared rays in order to detect the eye position of a face image, the structure of the system is simplified, and the extracted face image is eliminated. The eye position can be detected with high reliability using the histogram.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a conventional eye position detecting device.

FIG. 2 is a schematic block diagram of an eye position detecting device according to the present invention.

FIG. 3 is a schematic diagram of a face image including a search window according to the present invention.

FIG. 4 is a schematic diagram showing a search window including a pupil and a white eye shown in FIG. 3;

FIG. 5 is a schematic diagram showing a histogram of a search window in FIG. 4;

FIG. 6 is a flowchart illustrating an eye position detection method according to the present invention.

FIG. 7 is a flowchart illustrating an eye position detection method according to the present invention.

DESCRIPTION OF SYMBOLS 10, 100 Eye position detecting device 12 Infrared strobe 14 TV camera 16 A / D converter 18 Image memory 20 Eye window detecting unit 22 Iris detecting unit 24 Driver state determining unit 26 Timing circuit 110 Filter 120 Window generating unit 130 Histogram extraction unit 140 Peak detection unit 150 Area calculation unit 160 Switch 170 Selection unit 200 Face image 220 Search window 221 White eye 222 Skin 223 Pupil

Claims (10)

[Claims] 1. An eye position detection device for detecting an eye position in a face image, which is an image of a human face, each of which is represented by a plurality of pixels having a gray level value, wherein the eye position detection device surrounds the face image. Window generating means for generating a simple window; histogram extracting means for extracting a histogram from the window face image; peak detection for detecting a peak from the histogram and comparing the detected number of peaks with a predetermined number. Means, when the number of peaks matches the predetermined number, the histogram is a minimum pixel value including a peak value or a pixel number in a range between 0 pixel value points in a histogram for each of the window face images. Area calculating means for calculating an area, selecting the largest area among the calculated areas, and using the largest area Eye position detecting device which comprises a maximum area selection means for determining the position. 2. The eye position detecting apparatus according to claim 1, further comprising a filtering unit configured to filter the face image to remove noise in the face image to obtain a filtered face image. . 3. The eye position detecting device according to claim 2, wherein said filtering means is a median filter. 4. The eye position detecting device according to claim 1, wherein the predetermined number is three. 5. A window according to claim 1, wherein said window generated by said window generating means is a rectangle.
An eye position detecting device according to item 1. 6. The eye position detecting apparatus according to claim 5, wherein the window generating means scans the entire face image using a raster scanning method. 7. The histogram is obtained by plotting the frequency of pixels in the window face image as a function of gray level value, wherein the horizontal axis of the histogram represents gray level and the vertical axis is the 2. The eye position detecting device according to claim 1, wherein a frequency corresponding to each gray level is represented. 8. The eye position detecting device according to claim 1, wherein said peak detecting means regards a local maximum value of a histogram larger than a predetermined threshold value as a peak. 9. An eye position detection method apparatus for detecting, using a histogram, an eye position in a face image having a plurality of pixels each represented by a gray level value, the method comprising: filtering the face image; A first step of obtaining a filtered face image; a second step of generating a window in the filtered face image; a third step of extracting a histogram from the window; and a fourth step of detecting a peak of the histogram. A step of determining whether the number of the peaks is three, and determining the window as a candidate window if the number is three, a minimum peak among the three peaks of the candidate window A sixth step of determining two peaks having a maximum value and a maximum peak value; and providing the minimum peak value in the candidate window. That takes the area A _MAX indicating the number of pixels in the region surrounding the peak, the sum of the areas A _MIN representing the number of pixels in the region surrounding the peak with a peak value of said maximum, calculating the sum area A _T A seventh step of storing the sum area A _T and the corresponding position vector for the candidate window; and selecting the largest A _T among the stored sum areas A _T , Outputting a corresponding position vector as the position of the eye. 10. A case of detecting the positions of the two eyes, one of the sum area A _T, by selecting two largest sum area A _T, and outputs the position vector corresponding thereto as the position of both sides of the eye The eye position detecting method according to claim 9, wherein: